Re: [PATCH 04/13] mm: Fix pmd_read_atomic()

[Linus Torvalds]

On Sat, Oct 22, 2022 at 4:48 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> --- a/include/linux/pgtable.h
> +++ b/include/linux/pgtable.h
> @@ -258,6 +258,13 @@ static inline pte_t ptep_get(pte_t *ptep
>  }
>  #endif
>
> +#ifndef __HAVE_ARCH_PMDP_GET
> +static inline pmd_t pmdp_get(pmd_t *pmdp)
> +{
> +       return READ_ONCE(*pmdp);
> +}
> +#endif

What, what, what?

Where did that __HAVE_ARCH_PMDP_GET come from?

I'm not seeing it #define'd anywhere, and we _really_ shouldn't be
doing this any more.

Please just do

    #ifndef pmdp_get
    static inline pmd_t pmdp_get(pmd_t *pmdp)
    ..

and have the architectures that do their own pmdp_get(), just have that

   #define pmdp_get pmdp_get

to let the generic code know about it. Instead of making up a new
__HAVE_ARCH_XYZ name.

That "use the same name for testing" pattern means that it shows up
much nicer when grepping for "where does this come from", but also
means that you really never need to make up new names for "does this
exist".

             Linus